Prediksi Permintaan Daya Listrik Menggunakan Metode Long Short-Term Memory

Abstract

Electricity demand changes over time, making accurate prediction essential for effective energy planning and management. This research uses the Long Short Term Memory (LSTM) method to predict electricity demand in Singapore based on time series data, where the data are processed by forming input sequences with a certain number of timesteps so the model can learn patterns from past demand values before training. Hyperparameter tuning was carried out using Randomized SearchCV along with several data splitting schemes, and the results show that the best model performance was obtained using an 80% training and 20% testing data split with the Adam optimizer. Model performance was evaluated using Mean Absolute Percentage Error (MAPE) and the coefficient of determination (R square), where the optimal LSTM model achieved a MAPE value of 0.3223% and an R-square of 0.9966 on the testing data, indicating very high prediction accuracy and a strong fit between the predicted and actual electricity demand. Overall, these findings confirm that the LSTM method with an 80:20 data split and the Adam optimizer is effective for forecasting electricity demand in Singapore.